International Journal of Pattern Recognition and Artificial Intelligence

The incremental classifier is superior in saving significant computational cost by incremental learning on continuously increasing training data. However, existing classification algorithms are problematic when applied for incremental learning for multi-class classification. First, some algorithms, such as neural network and SVM, are not inexpensive for incremental learning due to their complex architectures. When applied for multi-class classification, the computational cost would rise dramatically when the class number increases. Second, existing incremental classification algorithms are usually based on a heuristic scheme and sensitive to the training data input order. In addition, in case the test instance is an outlier and belongs to none of the existing classes, few classification algorithms is able to detect it. Finally, the feature selection and weighing schemes being utilized are generally risky for a "siren pitfall" for multi-class classification tasks. To address the above problems, we bring forward an incremental gray relational analysis algorithm (IGRA). Experimental results showed that, when applied for incremental multi-class classification, IGRA is stable in output, robust to training data input order, superior in computational efficiency, and also capable of detecting outliers and alleviating the "siren pitfall".

This paper presents a structural distance-based crossover for neural network classifiers, which is applied as part of a Memetic Algorithm (MA) for evolving simultaneously the structure and weights of neural network models applied to multiclass problems. Previous researchers have shown that this simultaneous evolution is a way to avoid the noisy fitness evaluation. The MA incorporates a crossover operator that shows to be useful for ameliorating the permutation problem of the network representation (i.e. different genotypes can be used to represent the same neural network phenotype), increasing the structural diversity of the individuals and improving the accuracy of the results. Instead of a recombination probability, the crossover operator considers a similarity parameter (the minimum structural distance), which allows to maintain a trade-off between global and local search. The neural network models selected in this work are the product-unit neural networks (PUNNs), due to their increasing relevance in those classification problems which show a high order relationship between the input variables. The proposed MA is intended to reduce the possible overtraining problems which can raise in some datasets for this kind of models. The evolutionary system is applied to eight classification benchmarks and the results of an analysis of variance contrast (ANOVA) show the effectiveness of the structural-based crossover operator and the capacity of our algorithm to obtain evolved PUNNs with a higher classification accuracy than those obtained using other evolutionary techniques. On the other hand, the results obtained are compared with popular effective machine learning classification methods, resulting in a competitive performance.

Graphs provide us with a flexible and powerful way to represent objects in various areas of computer science. One of the main drawbacks is, however, that many standard algorithms on graphs have a high computational complexity. The present paper considers the problem of graph isomorphism, i.e. checking two graphs for identity. A novel approach for the efficient computation of graph isomorphism is presented. The proposed algorithm is based on bipartite graph matching by means of an assignment algorithm. The algorithmic framework is suboptimal in the sense of possibly rejecting pairs of graphs without making a decision. As an advantage, however, it offers polynomial runtime. In experiments on diverse graph data sets we demonstrate substantial speedups of our proposed method over several standard procedures for graph isomorphism. Furthermore, although the computational framework for isomorphism is suboptimal, we show that the proposed algorithm rejects only very few pairs of graphs and otherwise returns correct results.

Local learning has been successfully applied to transductive classification problems. In this paper, it is generalized to multi-class classification of transductive learning problems owing to its good classification ability. Meanwhile, there is essentially no ordinal meaning in class label of multi-class classification, and it belongs to discrete nominal variable. However, common binary series class label representation has the equal distance from one class to another, and it does not reflect the sparse and density relationship among classes distribution, so a learning and adjustable nominal class label representation method is presented. Experimental results on a set of benchmark multi-class datasets show the superiority of our algorithm.

Document image segmentation is an important research area of document image analysis which classifies the contents of a document image into a set of text and non-text classes. Previous existing methods are often designed to classify text and halftone therefore they perform poorly in classifying graphics, tables and circuit, etc. In this paper, we present a robust multi-level classification method using multi-layer perceptron (MLP) and support vector machine (SVM) to segment the texts from non-texts and thereafter classify them as tables, graphics and halftones. This method outperforms previously existing methods by overcoming various issues associated with the complexity of document images. Experimental results prove the effectiveness of our proposed method. By virtue of our multi-level classification approach, the text components, halftone components, graphic components and table components are accurately classified respectively which would highly improve OCR accuracy to reduce garbage symbols as well as increase compression ratio thereafter simultaneously.

This paper introduces a family of rectangularity measures. The measures depend on two parameters which enable their flexibility, i.e. the possibility to adapt with respect to a concrete application. Several rectangularity measures exist in the literature, and they are designed to evaluate numerically how much the shape considered differs from a perfect rectangle. None of these measures distinguishes rectangles whose edge ratios differ, i.e. they assume that all rectangles (including squares) have the same shape. Such property can be a disadvantage in applications. In this paper, we consider differently elongated rectangles to have different shapes, and propose a family of new rectangularity measures which assigns different values to rectangles whose edge ratios differ. The new rectangularity measures are invariant with respect to translation, rotation and scaling transformations. They range over the interval ]0, 1] and attain the value 1 only for perfect rectangles with a desired edge ratio.

In this paper, we propose a system for automatic image annotation that has two main components. The first component consists of a novel semi-supervised possibilistic clustering and feature weighting algorithm based on robust modeling of the generalized Dirichlet (GD) finite mixture. This algorithm is used to group image regions into prototypical region clusters that summarize the training data and can be used as the basis of annotating new test images. The constraints consist of pairs of image regions that should not be included in the same cluster. These constraints are deduced from the irrelevance of all concepts annotating the training images to help in guiding the clustering process. The second component of our system consists of a probabilistic model that relies on the possibilistic membership degrees, generated by the clustering algorithm, to annotate unlabeled images. The proposed system was implemented and tested on a data set that include thousands of images using four-fold cross validation.

Face recognition in constraint conditions is no longer a further challenge. However, even the best method is not able to cope with real world situations. In this paper, a robust method is proposed such that the performance of the face recognition system is still highly reliable even if the face undergoes large head rotation. Our proposed method considers local regions from half side of face rather than using the holistic face approach since in the former approach the "linearity" of features within the limited region is somewhat preserved regardless of the pose variation. Discrete wavelet transform is then utilized onto these patches in order to form face feature vectors. We train our recognizer using linear regression algorithm to interpret the relationship between a face vector for a specific pose and its corresponding frontal face feature vector. We demonstrate that our proposed method is able to recognize a non-frontal face with high accuracy even under low-resolution image by relying only on single frontal face in the database.

A novel system for verifying signatures on Persian handwritten bank checks is presented, in this paper. The presented system includes two main phases called: training and verification phases. At first, the system is trained using some genuine signatures provided by each customer in training phase. Then verifying the signatures on incoming checks is carried out in the verification phase. Feature extraction step is conducted based on a new approach that uses Multitresolution box-counting (MRBC) method for estimating the fractal dimension of signatures. Here, signature verification is modeled as testing hypothesis, and decision about acceptance or rejection of signatures on incoming checks is carried out using Kolmogorov–Smirnov test. The presented system has been tested on two databases: our new created database and NISDCC database which was used for ICDAR 2009 signature verification competition. Our database has 1000 genuine signatures provided by 100 participants and 200 skilled forgeries copied from genuine samples by five forgers. In total our database includes 1200 Persian signatures. Obtained results show promising performance of the presented system for its application on Persian banks.

Several 2DPCA-based face recognition algorithms have been proposed hoping to achieve the goal of improving recognition rate while mostly at the expense of computation cost. In this paper, an approach named SI2DPCA is proposed to not only reduce the computation cost but also increase recognition performance at the same time. The approach divides a whole face image into smaller sub-images to increase the weight of features for better feature extraction. Meanwhile, the computation cost that mainly comes from the heavy and complicated operations against matrices is reduced due to the smaller size of sub-images. The reduced amount of computation has been analyzed and the integrity of sub-images has been discussed thoroughly in the paper. The experiments have been conducted to make comparisons among several better-known approaches and SI2DPCA. The experimental results have demonstrated that the proposed approach works well on reaching the goals of reducing computation cost and improving recognition performance simultaneously.